Evidence Generation (LAERTES / CommoneEvidenceModel)

ericaVoss · November 21, 2017, 7:15pm

I was starting to read through the paper @rkboyce suggested:

Winnenburg R, Sorbello A, Ripple A, Harpaz R, Tonning J, Szarfman A, Francis
H, Bodenreider O. Leveraging MEDLINE indexing for pharmacovigilance - Inherent
limitations and mitigation strategies. J Biomed Inform. 2015 Oct;57:425-35. doi:
10.1016/j.jbi.2015.08.022. Epub 2015 Sep 2. PubMed PMID: 26342964; PubMed Central
PMCID: PMC4775467.

I’m starting to mark down what is different between Avillach and this method here in order to add this new method to our processing:

github.com

OHDSI/CommonEvidenceModel/blob/master/sandbox/MEDLINE_WINNENBURG.sql

/*
Winnenburg R, Sorbello A, Ripple A, Harpaz R, Tonning J, Szarfman A, Francis
H, Bodenreider O. Leveraging MEDLINE indexing for pharmacovigilance - Inherent
limitations and mitigation strategies. J Biomed Inform. 2015 Oct;57:425-35. doi: 
10.1016/j.jbi.2015.08.022. Epub 2015 Sep 2. PubMed PMID: 26342964; PubMed Central
PMCID: PMC4775467.
*/

/*Typically, ADEs are identified by the co-occurrence of a drug descriptor qualified by adverse effects and a disease descriptor qualified by chemically induced*/
/*Whils the method is generally similar to Avillach's, this tends to be more aggressive and goes beyond the simple co-occurrence of descriptors-qualifier pairs*/

/*IMPROVEMENT #1*/
/*2.1 The following MeSH terms imply denoting the chemically induced qualifier*/
SELECT *
FROM staging_medline.mesh_term
WHERE UI IN (
'D056486','D004893','D013262','D004409','D004409','D055963','D000014','D003875','D017109','D020230','D009459','D064807','D063926','D056487','D060831','D020267','D011605','D056150','D064146','D020258'
)
ORDER BY NAME

This file has been truncated. show original

If anyone already has experience with this let me know. Otherwise this is just a shout out to @rkboyce, @anthonysena, @pnatsiavas, @vkoutkias, @abedtash_hamed, & @lee_evans that I was starting to think about this.

rkboyce · November 23, 2017, 6:49am

Erica - Here is the actual data from July 2017 - https://drive.google.com/open?id=0B7zaa0UrHMiaSVlDZExLWG50M28

I can ask Olivier at NLM for a new dump when you are ready. So, could you develop the load strategy using the actual data?

Happy Thanksgiving,
-Rich

schuemie · November 23, 2017, 8:08am

Thanks @rkboyce!

Since our goal is to create a process that can update the data on a continuous basis, I would recommend we re-engineer Olivier’s process so we can automatically execute it at every refresh rather than add a dependency on Olivier to run it every time.

We can use the actual data you provided to test our implementation.

Christian_Reich · November 23, 2017, 3:53pm

What are you guys streaming from Olivier?

ericaVoss · November 29, 2017, 2:13pm

@schuemie & @rkboyce,

I agree, let’s try to recreate the process into a format that works with our OHDSI framework. I’m having it loaded internally now so that I can see what is going on. Already see a few things that I didn’t take away from the article so I’m looking into that more. For example, why are tags associated to “precoordinated” pulled and what is being done with indications?

Thank you for sharing this @rkboyce!

@Christian_Reich - we are trying to recreate this process and Rich got a copy of the output directly from the source.

Winnenburg R, Sorbello A, Ripple A, Harpaz R, Tonning J, Szarfman A, Francis
H, Bodenreider O. Leveraging MEDLINE indexing for pharmacovigilance - Inherent
limitations and mitigation strategies. J Biomed Inform. 2015 Oct;57:425-35. doi:
10.1016/j.jbi.2015.08.022. Epub 2015 Sep 2. PubMed PMID: 26342964; PubMed Central
PMCID: PMC4775467.

rkboyce · November 29, 2017, 2:33pm

Sounds good. Let’s do it! Will help and we can always ask questions of
folks at NLM.

Christian - this is data that NLM created that improves upon the
Avillach method for identifying PubMed titles and abstracts about
drug-HOI associations using MeSH indexing.

ericaVoss · December 4, 2017, 4:30pm

@rkboyce - Do you know of any other documentation other than the README? I’m asking because it looks like they did things outside of the paper as well as I don’t know what some of their columns mean (DRUGS.DRUG_ROLE what is i and c).

rkboyce · December 4, 2017, 9:24pm

Will check using a separate email CCing you…

Christian_Reich · December 14, 2017, 5:16pm

Friends:

Sounds like cool stuff. Is there anything required from me? How can I help?